Reading your results

Luria shows you the same numbers it uses to decide. This page reads them with you, field by field, and tells you when a number is worth trusting.

What you'll need

• Luria live on your site with traffic flowing. See Verify your install.
• At least one test running. See How Luria's tests work.

The Split Tests card

Each running test gets one card.

• Name and where. What the test changes, and the page it runs on.
• Badge. "running", "paused", or "ready to promote" once the test has cleared the promotion bar.
• control and variant bars. The conversion rate measured on each arm, to one decimal. Sessions counted here are sessions that were actually shown that arm.
• {n} sessions. Both arms added together. Traffic stays split evenly until each arm alone is past 200.
• P(beat): {x}%. The probability that the variant is genuinely better than control, given the counts so far. Above 95% is the promotion bar. Below 5% and the variant gets killed.
• {x}pp, estimate. The gap between the two arms in percentage points, as a middle estimate rather than a fact. One percentage point on a 2% conversion rate is a 50% relative change, so read pp carefully.

The heading above the cards reads "Running now, next split Thu 3:00 AM". That line is fixed text in the dashboard, not a schedule for your store. Allocation is recalculated every hour.

Two sections lower down, "Planned for the next split" and "Multi-product testing", render empty today.

The Overview cards

• Conversion rate, with "measured on your tracked traffic" underneath. Conversions divided by the sessions Luria saw. It will not match Shopify exactly; see Numbers don't match Shopify.
• Lift, with one of two labels. "Lift since start, measured" means your 4-week baseline is frozen and the number is a real comparison. "Lift, estimate (baseline not frozen yet)" means Luria does not have four weeks of data yet, so the figure is directional only. Both are shown against your 4-week baseline.
• Conversions this period, with tracked revenue underneath when your goal is Sales. Revenue is net of fully refunded orders.
• Active tests. How many tests are running right now.
• Conversion rate, last 12 weeks. Weekly conversion rate built from your own events, labelled with week 1, this week, and a note that a circle marks a promoted test.

Why the numbers move early

Early numbers swing because the samples are small, not because something is broken.

• Under 200 sessions per arm the split is held even on purpose, and P(beat) can wander from 20% to 80% on a handful of conversions.
• A single large order moves tracked revenue much more than it moves the conversion rate.
• The lift card stays on "estimate" until you have four weeks of data behind the baseline.

When to trust a number

• Both arms are past 200 sessions. That is the bar Luria promotes on, and it is the right bar for you too.
• P(beat) is above 95% or below 5% and holding, rather than crossing back and forth day to day.
• The card badge says "ready to promote". That means the sample bar and the 95% bar were both cleared.
• The lift card says "measured", not "estimate". Anything labelled estimate is a modelled range, not a measurement.

Common failures

• A card shows 0 sessions. The change is live but no visitor has been shown it yet, or serving is off. See No variants serving.
• Lift never leaves "estimate". The baseline needs four weeks of data. Low traffic stretches every window.
• Conversions look too low. Luria counts attributed orders once each, net of full refunds. See Orders missing and Duplicates.
• P(beat) sits at 50%. The two arms are behaving the same. That is a real result, and Luria will keep the test running until the gap is decidable.

Next steps

• How a test is measured
• How Luria's tests work
• Your dashboard
• Numbers don't match Shopify